Claude Code adds an automated permission determination feature, where AI checks permission risks and automatically executes tasks.



Claude Code, a coding assistant, has added an ' Auto mode ' where AI makes permission decisions on behalf of the user. This feature automates certain tasks that require human approval.

Auto mode for Claude Code | Claude

https://claude.com/blog/auto-mode




Auto mode is a feature where AI evaluates each action of each tool to check for any potentially dangerous actions such as mass file deletion, leakage of sensitive data, or execution of malicious code. Actions that the AI deems safe are executed automatically, while risky actions are blocked, prompting the user to take an alternative approach. If Claude continues to take actions that are blocked, the user will eventually be prompted for permission.

According to its developer, Anthropic, Claude Code's default permission settings are intentionally conservative, requiring approval for all file writes and bash commands, for example. Because of these settings, it cannot automate large-scale tasks entirely without human intervention.

However, some developers use the '--dangerously-skip-permissions' command, which automates the process by eliminating permission approval. Anthropic points out that 'while this command bypasses permission checks, it can lead to dangerous and uncomfortable consequences and should not be used outside of isolated environments,' and recommends Auto mode as an intermediate approach.



While Auto mode reduces risk compared to --dangerously-skip-permissions, it is not entirely risk-free, and its use in isolated environments is still recommended.

Auto mode is available as a research preview for the 'Team' plan. It will be available to Enterprise plan and API users in the coming days.

Related Posts:

in AI, Posted by log1p_kr